Staff ML Scientist

Visa Visa · Fintech · Bengaluru, India, IN

Staff ML Scientist at Visa in Bengaluru, India, focusing on designing and developing next-generation AI/ML models for the payments ecosystem, specifically in transaction monitoring, fraud detection, and risk mitigation. The role involves fine-tuning LLMs, framing classification problems as generative tasks, leading ML PoCs for agentic commerce, and integrating models into production. Emphasizes research, innovation, and mentoring.

What you'd actually do

  1. Extract insights from large-scale Visa datasets across business lines to define high-value, fit-for-purpose problems beyond payments - driving clarity and framing business impact with precision.
  2. Fine-tuning open-weight LLMs to reduce cost – bring deep expertise in parameter-efficient fine-tuning (LoRA/QLoRA) and distillation to achieve near frontier performance at significantly lower inference and training cost.
  3. Classifier as an LLM model - frame classification problems (e.g., fraud signals, intent detection) as generative tasks using LLMs, enabling better generalization and faster iteration vs. traditional pipelines.
  4. Lead other ML PoCs - risk scoring of stablecoins, intent misalignment, and merchant discoverability for agentic commerce. Rapidly prototype and validate new problem spaces with lightweight experimentation frameworks to identify high-ROI opportunities for scaling.
  5. Rapidly design, experiment, and deploy SOTA ML/AI models, owning the full lifecycle and ensuring solutions are scalable, measurable, and aligned with strategic goals.

Skills

Required

  • 5+ years of relevant work experience with a Bachelor’s Degree or at least 2 years of work experience with an Advanced degree (e.g. Masters, MBA, JD, MD) or 0 years of work experience with a PhD, OR 8+ years of relevant work experience.
  • Strong fundamentals in machine learning, deep learning, and optimization and solid understanding of the math behind them.
  • Ability to derive algorithms from first principles and understand their limitations.
  • Proven track record (6+ years) of building AI / ML models which can handle scale (think model size, latency, architecture, etc).
  • Shipped models that matter - ideally in production environments with real users and measurable impact.
  • Deep technical and domain expertise in at least two domains where you are able to connect different ML techniques (e.g. contrastive learning, domain adaptation, etc.) to appropriate business problems.
  • Fluency in code.
  • Write clean, efficient Python, and comfortable navigating large codebases.
  • Treat code as a medium for clarity and experimentation.
  • Experience in PySpark, PyTorch, tensor flow, as well as GPU based compute environments.
  • Research mindset with product intuition.
  • Ability to balance novelty with practicality and know how to translate ideas into working systems.
  • Deeply technical covering both ML as well as deep learning fundamentals spanning design, pre-training, fine-tuning, and evaluation.
  • Ownership and velocity.
  • Ability to take initiative, move fast, and iterate.
  • Ability to define the problem, build the solution, and drive it to completion.
  • Communication clarity.
  • Ability to explain complex ideas simply, whether in a design doc, a pull request, or a hallway conversation.
  • Write well and think clearly.

Nice to have

  • PhD/MS in computer science, machine learning or related field preferred.

What the JD emphasized

  • Proven track record (6+ years) of building AI / ML models which can handle scale (think model size, latency, architecture, etc). You’ve shipped models that matter - ideally in production environments with real users and measurable impact.
  • deep technical and domain expertise in at least two domains where you are able to connect different ML techniques (e.g. contrastive learning, domain adaptation, etc.) to appropriate business problems.
  • You write clean, efficient Python, and you’re comfortable navigating large codebases.
  • You treat code as a medium for clarity and experimentation.
  • You have experience in PySpark, PyTorch, tensor flow, as well as GPU based compute environments.
  • You know when to read papers and when to write them.
  • You can balance novelty with practicality and know how to translate ideas into working systems.
  • You are deeply technical covering both ML as well as deep learning fundamentals spanning design, pre-training, fine-tuning, and evaluation.
  • Ownership and velocity. You take initiative, move fast, and iterate.
  • You don’t wait for permission, you define the problem, build the solution, and drive it to completion.
  • Communication clarity. You can explain complex ideas simply, whether in a design doc, a pull request, or a hallway conversation.
  • You write well and think clearly.

Other signals

  • design and develop next-generation AI and ML models
  • solve real-world challenges in the payments ecosystem
  • driving innovation in the Risk and Identity Solutions (RaIS) through advanced AI/ML technologies
  • drive the scientific roadmap at scale
  • models are aligned, explainable, and robust
  • influencing product strategy in a high-stakes domain
  • top-tier compute, tools, and datasets
  • fine-tuning open-weight LLMs to reduce cost
  • parameter-efficient fine-tuning (LoRA/QLoRA) and distillation
  • Classifier as an LLM model
  • risk scoring of stablecoins, intent misalignment, and merchant discoverability for agentic commerce
  • Rapidly prototype and validate new problem spaces
  • Rapidly design, experiment, and deploy SOTA ML/AI models
  • owning the full lifecycle
  • Lead future-forward AI/ML research
  • publishing influential work, and filing patents
  • Integrate models into production
  • building robust, compliant pipelines
  • Mentor scientists using Generative AI tools